If you’re spending fifteen minutes rewriting every letter your AI scribe produces because it just doesn’t sound like you, a few settings changes won’t properly fix it. The problem goes deeper than that - you’re working around a model that wasn’t built for your specialty or for it to sound like you. Here’s the short version of why that happens and what actually fixes it.
The short version
The output isn’t wrong in the way you might expect. The facts are usually there. The structure is roughly correct. But the clinical emphasis is off. The terminology feels like a registrar covering a rotation they haven’t done before. The letter could have come from any doctor in any specialty. It doesn’t sound like you, because the AI genuinely doesn’t know that you’re a specialist.
A clinician recently reached out to Medow’s team, saying they found us because they searched: "Heidi doesn’t sound like me, are there any scribes set up for specialists?" That question captures the problem exactly. The doctor wasn’t asking how to fix templates. They were asking whether a different tool existed that actually sounded like them. The answer is yes. The same frustration can apply whether you’re using Heidi, Lyrebird, Dragon Medical One, i-Scribe, or any other tool built for a general clinical audience. The reason it happens is the same in every case, and it starts with how some of these models are built.
Most doctors who switch away from a general-purpose AI scribe describe the same experience in one of three ways. You might recognise your own.
The wrong things get emphasised
The AI captures everything said in the consult and gives it roughly equal weight. A passing comment about fatigue ends up foregrounded the same way a haemodynamic finding does. A relevant surgical history gets buried. The letter reads like a transcript of the conversation, not like a clinical assessment from someone who understands which findings are significant in your type of practice.
The terminology is slightly off
The drug names are close but not right. The procedure names are generic rather than specialty-specific. The abbreviations your field uses are spelled out, or substituted with equivalents from another discipline. Nothing is wrong enough to be dangerous, but every letter needs to be edited before it leaves the practice, which erases most of the time saving.
The structure doesn’t match your specialty’s conventions
A cardiologist’s letter to a GP is structured differently from a gastroenterologist’s. A psychiatrist’s assessment has a different register and level of detail from a post-operative surgical note. A pain specialist frames functional findings in a specific way that a general model doesn’t replicate. The output structure the AI defaults to is somewhere between a SOAP note and a GP summary, because that’s the most common format in the training data.
These three problems have the same root cause. And it isn’t your settings. You can read more about how this plays out across specific specialties in Medow’s guide to AI scribes for specialist clinics in Australia.
General-purpose AI scribes are trained on broad clinical data spanning hundreds of specialties: GP notes, allied health records, emergency presentations, hospital discharge summaries, and everything in between. The model learns to produce output that is statistically average across all of that clinical writing.
The result is output that resembles no specialty in particular. It is technically coherent. It captures the facts. But it produces letters and notes in a register that belongs to no one, because it was built to serve everyone.
This is a model architecture problem more so than a configuration problem. When a general model generates a specialist letter, the errors it makes tend to be errors of omission rather than errors of invention: it leaves out the specialty-specific detail that makes a letter clinically useful, because it was never trained to know what that detail was. For a specialist, those omissions cluster in exactly the sections that matter most to the receiving GP.
No amount of template adjustment teaches a general model which clinical details are significant in your specialty, which abbreviations are standard in your field, or how referral letters from your discipline are conventionally structured. The model architecture determines the ceiling. Templates work within that ceiling. They can’t raise it.
The standard workaround offered by general-purpose scribes is template customisation. Build your own template, they say, and the output will match your style.
Template customisation changes structure. It does not change clinical intelligence. A general model with a custom template still won’t understand which medications are relevant to document in your specialty’s context, which symptoms warrant a dedicated section in your type of report, or how a specialist in your field frames the management plan for a referring GP.
There is also a more fundamental problem: this approach puts the burden on the doctor. You are expected to become, in effect, an AI engineer for your own clinical tool, building and refining templates until the output is acceptable. Most specialist doctors do not have the time or inclination to do that. And they shouldn’t have to.
The promise of an AI medical scribe is that it saves you time. The real-world outcomes from Australian specialist clinics using Medow make that concrete: an ENT surgeon saving over 20 hours every month in admin, and a WA ophthalmology practice reclaiming 10 hours every week. That return disappears entirely if you are spending the time you saved on template engineering instead.
Getting a letter or note to sound like a specific specialist doctor requires three layers working together. Each one needs to be built in, not bolted on.
1. A specialist medical dictionary for your field
The correct drug names, procedure names, anatomical references, and abbreviations for your specialty, including Australian medication names, hospital names, and local clinical conventions. Medow builds a dedicated medical dictionary for each specialty, covering terminology and localisation. You can see the full list of supported specialties at medowhealth.ai/specialties.
2. Clinical relevance filtering tuned to your specialty
Not every detail from a consult belongs in every type of report. A cardiologist’s letter to a GP handles haemodynamic findings differently from how a geriatrician frames a functional decline assessment. A psychiatrist’s note captures elements of the therapeutic encounter that a surgical post-op note would never include. Medow’s specialist models are trained on the information hierarchy of each specialty: what matters, what doesn’t, and how to weight it in the generated output.
3. Your individual voice and writing style
Beyond specialty conventions, every doctor has their own clinical voice: the level of formality, the degree of detail in the history section, preferences around sentence length, the particular phrases you use to frame a management plan. Medow profiles your voice and personalised it before you go live. You upload three of your existing letters and Medow builds a Voice Profile from them. From that point, every report and letter the AI generates reflects your individual writing style, not just the average style for your specialty. The result is output that is "Unmistakably you" - which is exactly what Medow’s campaign line promises, because it is an accurate description of what the feature delivers.
"Most of the letters we hardly need to check now, they sound exactly like we wrote them. I only make minor adjustments, if needed at all."
.png)
Medow is a Specialist-grade AI medical scribe built specifically for specialist clinics in Australia and New Zealand. Specialist doctors who switch to Medow from a general-purpose tool tend to describe the same moment: they open the first letter and find themselves reading it to check it says what they would say, rather than reading it to find what needs fixing. That shift in posture (from correction to confirmation) is the practical difference a specialist model makes. Rather than a single shared model adapted with templates, Medow maintains independent AI models for each medical specialty, currently covering more than 50 specialties. Each model is trained from the ground up on the medical knowledge, clinical context, and information hierarchy of that specialty, alongside clinical and AI experts. You can see a full comparison of how Medow differs from Heidi and i-Scribe in the Medow vs Heidi vs i-Scribe comparison.
The distinction matters because it means the model understands your specialty’s conventions before you ever log in, not after weeks of template adjustments. Cardiology for cardiologists. Psychiatry for psychiatrists. Gastroenterology for gastroenterologists. Each model has its own specialist dictionary, its own clinical relevance framework, and its own understanding of what a report in your specialty is expected to contain.
Medow also includes verification checks on every output. All generated letters and notes are checked against the transcript of the consult, preventing the AI from adding clinical conclusions or diagnoses that were not provided by the doctor. For specialist practice, where a hallucinated finding in a referral letter carries real clinical and medico-legal weight, this is a meaningful safety feature.
"After trialling a number of AI scribes, Medow clearly stands out. The output quality is consistently excellent, the customisation capabilities are advanced and flexible, and the support from our dedicated account manager has been exceptional. Altogether, it's given me meaningful time back each day."
There is a pattern that plays out in almost every specialist clinic that has tried a general-purpose AI scribe. The doctor or their staff spend days, sometimes weeks, adjusting templates and settings trying to get the output close enough to use. They tweak the structure. They add vocabulary. They correct the same errors repeatedly. Each time it gets slightly better. But it never quite gets there, and because it never quite gets there, the doctor never quite trusts it. They still read every letter in full before signing, because they have been burned before. The time saving the tool promised becomes theoretical.
This is the cost that never shows up in a ROI calculator: the hours of staff time spent on configuration, the doctor’s cognitive overhead of reviewing every output with suspicion, and the slow erosion of confidence that makes adoption fail even when the underlying technology is sound. For a busy specialist clinic, that cost is real and it compounds every week.
A specialist AI model gets the clinical conventions right. What it cannot do on its own is sound like you specifically from day one, without input on your individual writing style. That is the gap the Customer Success Manager closes, before you ever see your first output. And because Medow’s approach removes the configuration burden from the doctor entirely, the trust problem never develops in the first place.
Before you go live, a dedicated Customer Success Manager configures your instance of Medow. They take samples of your existing letters and reports and use them to set your formatting preferences, layout, and tone, all before you have access to the platform. You do not need to build templates. You do not need to spend your first weeks editing outputs into shape. The scribe is calibrated to your voice before your first consult.
After onboarding, the Customer Success Manager continues to work with you. Edge cases get resolved. New customisations get added. If you change how you want a particular section structured, or if a new report type needs to be added, the CSM handles it. Feedback you give flows back into the specialist AI model, contributing to ongoing improvements.
Critically, unlike self-serve AI scribes, the doctor is not expected to do any of the setup or tuning. That work belongs to the Customer Success team. For clinics switching from an existing tool, the CSM also handles the migration: replicating any conventions you had already configured elsewhere, so the transition does not feel like starting from scratch. This is also what made the transition from a dictation service straightforward for clinics like the practice featured in Medow’s medical dictation guide: the CSM handled the configuration that would otherwise have fallen on the practice.
"Medow has been a huge time-saver for my practice, and the quality of the letters is excellent. What really stands out to me is how much the Medow team cares about ongoing product improvement. They actively listen to feedback, work closely with customers, and continually optimise and individualise the system to suit our needs. I’ve found that level of engagement and responsiveness very impressive."
"Medow Health has exceeded my expectations. It captures my consults accurately, learns my style, integrates with Gentu, and significantly reduces my paperwork. With hands-on support, adopting the system has been even easier!"
If you are considering switching from your current scribe, these three questions will help you cut through the marketing claims quickly.
A dedicated Customer Success Manager who configures the system before go-live and continues refining it afterwards eliminates that pattern entirely. You trust the output from day one because someone who knows the system built it for you before you ever logged in. If you’re evaluating the broader landscape, our guide to the best AI medical scribes in Australia applies these criteria across the leading tools.
Most AI medical scribes use a single general-purpose model trained on clinical data from across hundreds of specialties and settings. The model produces output that is average across all of that clinical writing, which means it does not reflect the specific conventions, terminology, or clinical register of any individual specialty. For specialist doctors, this is particularly noticeable because specialist practice has its own distinct vocabulary, report structures, and clinical emphasis that a general model never learns. The errors that result tend to be errors of omission: the model leaves out specialty-specific detail rather than inventing facts, because it was never trained to recognise what that detail was in the first place.
Template customisation can improve the structure of the output, but it does not change the underlying clinical intelligence of the model. A general model with a custom template will still lack knowledge of which clinical details are significant in your specialty, which abbreviations are standard in your field, and how reports in your discipline are conventionally written. Template customisation also places the configuration burden on the doctor, which removes much of the time-saving benefit the scribe was supposed to deliver.
A specialist AI medical scribe like Medow uses independent AI models trained specifically for each medical specialty, rather than a single shared model adapted with templates. Each specialist model is trained on the medical knowledge, clinical context, terminology, and information hierarchy of that specialty. This means the AI understands your specialty’s conventions from day one: which findings are significant, which abbreviations are standard, and how letters and notes in your field are expected to read.
Heidi is a general-purpose AI scribe designed to serve a wide range of clinicians across all specialties and settings. Medow is built specifically for specialist clinics, with independent AI models for each specialty rather than a shared model with templates. Medow also assigns a dedicated Customer Success Manager to each clinic who configures the system before go-live using samples of the doctor’s existing letters, and continues refining it afterwards. For specialist doctors who have tried Heidi and found the output too generic for their practice, the difference is structural rather than superficial: the underlying model is different, not just the settings, which is why the output quality change is immediate rather than gradual. You can compare the two tools in detail in the Heidi vs i-Scribe vs Medow comparison.
From day one, rather than over time. Before you access the platform, Medow’s Customer Success Manager uses samples of your existing letters and reports to configure your Voice Profile and set your preferred formatting, layout, and tone. You do not need to spend weeks editing outputs into shape. The system is calibrated to your individual clinical voice before your first consult. Ongoing refinement continues through the CSM relationship after go-live.
With self-serve AI scribes, setup and ongoing tuning typically falls to the doctor or their practice staff. In busy specialist clinics, this means days or weeks of configuration time before the output reaches an acceptable standard. Even then, persistent gaps in quality mean the doctor reviews every letter before signing, which eliminates much of the time saving. With Medow, a dedicated Customer Success Manager handles all configuration before go-live, using samples of your existing letters to calibrate the system to your voice. The doctor’s staff do not need to become AI engineers. The doctor does not need to rebuild trust slowly over weeks of inconsistent outputs. The system works correctly from the first consult.
Medow supports more than 50 medical specialties, with dedicated AI models for each. Supported specialties include cardiology, dermatology, ENT, gastroenterology, general surgery, geriatrics, neurology, neurosurgery, orthopaedics, psychiatry, urology, and many others. See the full list at medowhealth.ai/specialties.
Yes. Dragon Medical and traditional dictation services have a different version of the same problem: they capture what you say verbatim and format it, but they have no clinical intelligence about your specialty. The output reflects your words but not your specialty’s conventions, and the burden of structuring a letter that reads correctly still falls on you or your staff. Medow’s AI Scribe and Smart Dictation both replace this workflow, with the specialist model and Voice Profile ensuring the output already reflects the structure and terminology your specialty expects, without you having to dictate around a template. If you’re currently using Dragon or a typist and considering moving to AI, the Dragon Medical vs Medow comparison covers the practical differences in detail.
